Digital Humans Deep Dive

Complete reference for Digital Human fields, generation, and best practices

This page serves as the detailed reference for Digital Humans on Bluejay. For a conceptual overview, start with the Digital Humans Overview.

What Is a Digital Human?

A Digital Human is a synthetic replica of a customer that interacts with your agent during Bluejay simulations. Each Digital Human carries an intent, success criteria, and a set of behavioral traits that together form a single, self-contained test case.

Digital Humans are the atomic unit of testing on Bluejay. A simulation runs one or more Digital Humans against your agent, producing a transcript and evaluation result for each interaction.

Digital Human Fields

Every Digital Human is defined by the following fields:

Field	Type	Description
Intent	String	What the Digital Human wants to accomplish in the conversation
Success Criteria	List	Conditions that must be met for the interaction to be considered successful
Language	String	The language the Digital Human speaks (e.g., English, Spanish, Mandarin)
Accent	String	Regional accent applied to speech (e.g., British, Southern US, Latin American)
Emotion	String	Emotional tone (e.g., Calm, Frustrated, Anxious, Angry, Cheerful)
Speaking Speed	String	Pace of speech (Slow, Normal, Fast)
Volume	String	How loud the Digital Human speaks (Quiet, Normal, Loud)
Background Noise	String	Ambient sounds during the call (Office, Airport, Car, None)
Scripted Responses	Map	Trigger-response pairs for deterministic behavior
DTMF Sequences	List	Touch-tone codes to send during the call
Silence Duration	Number	Seconds the Digital Human stays silent at a specified point
Allow silence tool	Boolean	When true, the voice runtime may use the silence tool for this digital human (subject to execution-layer rules)
Silence tool instructions	String	Use "default" for built-in silence-tool behavior; otherwise custom instructions for the runtime. Distinct from scripted silence duration above

How Bluejay Generates Digital Humans

When you use the generation endpoint, Bluejay creates Digital Humans based on the context you’ve provided about your agent. The generation process works in three stages:

1. Context ingestion

   Bluejay reads your agent’s description, system prompt, goals, and any existing Digital Humans to understand the problem space.

2. Scenario diversification

   The generation engine creates a diverse set of intents and customer profiles, varying across languages, emotions, complexity levels, and scenario types.

3. Criteria assignment

   Each generated Digital Human receives tailored success criteria aligned to its specific intent and your agent’s expected behavior.

The more you tell Bluejay about your agent’s behavior, the better the generation engine can produce diverse, relevant Digital Humans. Include edge cases, known failure modes, and multi-step workflows in your agent description.

Intent Design

A well-written intent is specific, actionable, and reflects a real customer scenario. Here’s how to think about writing intents:

Weak intent:

"Ask about billing"

Strong intent:

"I was charged $49.99 on March 3rd but I cancelled my subscription on February 28th.
I want a full refund and written confirmation that no further charges will occur."

Strong intents give the Digital Human a clear objective and provide enough context for the conversation to feel authentic. They also make success criteria easier to define because the expected outcome is already implied.

Success Criteria Design

Success criteria should be specific, observable, and binary. Each criterion should be something Bluejay can evaluate from the transcript alone.

Good criteria:

• The agent acknowledged the duplicate charge within the first 3 turns
• The agent offered a refund or escalated to billing
• The agent did not share account balance before verifying identity

Bad criteria:

• The agent was helpful
• The conversation went well
• The agent sounded professional

Scripted Responses vs. Natural Conversation

Digital Humans support two conversation modes:

Natural conversation

The Digital Human responds dynamically based on its intent and traits. Best for exploratory testing and discovering unexpected agent behaviors.

Scripted responses

The Digital Human follows a predetermined script, responding with specific phrases when triggered. Best for regression testing and deterministic validation.

You can mix both modes — a Digital Human can follow a script for the first few turns (providing account details, answering verification questions) and then switch to natural conversation for the resolution phase.

Communities

Digital Humans can be grouped into Communities — reusable sets of personas that form a benchmark population. Communities enable:

• Consistent benchmarks — run the same set of Digital Humans against different agent versions to compare performance over time
• Audience segmentation — organize Digital Humans by customer type, language, or scenario category
• Cross-agent testing — use one Community across multiple agents to see how different agents handle the same customers

See the Communities documentation for details on creating and managing Communities.

Best Practices

1. Start with your real customers

   Review your production call logs, support tickets, and customer feedback. Build Digital Humans that reflect the actual scenarios your agent encounters.

2. Cover the full spectrum

   Don’t only test the happy path. Create Digital Humans for edge cases, adversarial scenarios, multilingual interactions, and silence/timeout conditions.

3. Be specific with success criteria

   Vague criteria produce vague results. Write criteria that are observable in a transcript and leave no room for interpretation.

4. Iterate and expand

   Start with a small set of targeted Digital Humans. As you discover new failure modes in production, add corresponding Digital Humans to your simulation suite.

5. Use generation for breadth, manual for depth

   Generate large populations to discover unexpected scenarios. Then create hand-crafted Digital Humans for your most critical and nuanced test cases.

API Reference

Create

Create a single Digital Human with full configuration.

Generate

Auto-generate up to 100 diverse Digital Humans.

Update

Modify an existing Digital Human’s fields.